Liberal vs Restrictive Hemoglobin Levels as Thresholds for Blood Transfusions

Thresholds had similar 30-day mortality; restrictive threshold results in fewer transfusions

Benefits in NNT

No one was helped (no difference in mortality rate between the two strategies)
1 in 3 were helped (transfusion avoided with a restrictive strategy)
No one was helped (no difference in mortality rate between the two strategies)
36% lower risk of blood transfusion with a restrictive strategy

Harms in NNT

No one was harmed
No one was harmed
View As:

Efficacy Endpoints

30-day mortality

Harm Endpoints

Cardiac events, myocardial infarction, congestive heart failure, rebleeding, sepsis, pneumonia, thromboembolism, renal failure


Blood transfusion is a common treatment of anemia due to chronic disease or acute blood loss.1 However, there continues to be uncertainty concerning the appropriate threshold for transfusion. A restrictive protocol could decrease blood administered, transmissible infections, transfusion reactions, volume overload, and utilization of a limited commodity. However, anemia may result in decreased oxygen delivery which could lead to metabolic dysfunction and increased cell death. Determining an appropriate transfusion threshold is thus an important objective. A previous Cochrane review2 found the use of restrictive thresholds decreased transfusions compared to liberal thresholds. No difference in secondary outcomes, including 30-day mortality, hospital length of stay, cardiac events, and myocardial infarction, was noted.

Here we summarize an updated Cochrane review of 31 trials and 12,587 adult participants.3 Included studies were randomized trials that assigned patients to either liberal or restrictive transfusion thresholds, typically defined as a hemoglobin level of 9-10 g/dL or 7-8 g/dL, respectively. The primary outcome was 30-day mortality. Secondary outcomes included the proportion of participants transfused, number of units transfused, cardiac events, non-fatal and fatal myocardial infarction, renal failure, and congestive heart failure.

The authors found no difference in 30-day mortality between liberal and restrictive transfusion protocols (Relative risk [RR] 0.97, 95% CI 0.8 to 1.2). However, 84% of the subjects assigned to liberal thresholds received a blood transfusion, compared to 48% of the restrictive group subjects, an absolute reduction of 36% (NNT of 3 for avoiding red cell transfusion). The liberal threshold also resulted in 1.3 more units being transfused per participant than the restrictive threshold. There was no difference between the two groups in cardiac events, myocardial infarction, congestive heart failure, rebleeding, sepsis, pneumonia, thromboembolism, or renal failure.

Subgroup analysis of 30-day mortality was also performed and two subgroup results are of particular note. The first consisted of patients with an acute myocardial infarction, for whom two studies found 30-day mortality was numerically (though not statistically) higher in the restrictive than in the liberal group, with low numbers of deaths (9/78 vs. 2/76) and low numbers overall in the analysis (n=154). On the other hand, participants with acute gastrointestinal bleeding in three studies had a significantly lower mortality with the restrictive protocol than with the liberal protocol [RR 0.65 CI 0.43-0.97, n=1522, NNT 37].


This Cochrane review was a comprehensive and important update to the prior review, but several important caveats should be kept in mind. These include the potential for biases due to lack of blinding, statistical heterogeneity in the secondary outcomes, and differences among study methodologies. The intervention examined in this study, the administration of blood, makes blinding difficult, increasing risk of bias as only one trial blinded participants to treatment. Another concern is statistical heterogeneity. While the primary outcome had low heterogeneity, some secondary outcomes, including the risk of receiving blood transfusion, showed heterogeneity. Finally, important methodological differences between the studies could have affected the validity of the findings. One important example is differing restrictive transfusion thresholds.

Half of the studies used a restrictive threshold of 7g/dL, while the other half used 8g/dL. A transfusion threshold of 7 g/dL was more common in studies of ICU patients. Study populations varied widely, and different transfusion thresholds may affect different patients and conditions differently. This is apparent in the subgroup analysis where there were signals of possible benefit and harms. Subgroup analysis of patients with acute gastrointestinal bleeding demonstrated benefit with the restrictive protocol, while analysis of patients with acute myocardial infarction demonstrated possible harm. The implication of these findings is unclear, though they suggest areas for research.

While there was no difference in mortality or adverse outcomes, restrictive transfusion criteria did decrease transfusions, a potential benefit we see as important based on resource utilization and harm risks associated with transfusion.

Of note, a systematic review published in 2020 evaluated the effects of a restrictive compared to a liberal hemoglobin threshold in intensive care patients, and supports the findings of the Cochrane review.4 Eight randomized controlled trials and 3415 subjects were included and no significant difference was found in short-term mortality, length of hospital stay, length of ICU stay, or ischemic events.

As there was no evidence suggesting benefits with the routine use of a liberal transfusion threshold and since a liberal transfusion threshold leads to greater blood administration, we have assigned a color recommendation of Black (harms > benefits). In summary, routine use of a liberal hemoglobin threshold demonstrated no clear medical benefit or harm in regards to 30- day mortality and morbidity outcomes. However, the liberal threshold did lead to increase utilization of blood products, which could pose a risk of infection and transfusion reaction.

The original manuscript was published in Academic Emergency Medicine as part of the partnership between and AEM.


John Z. Hillenkamp, MD; Allan B. Wolfson, MD
Supervising Editors: Shahriar Zehtabchi, MD


October 15, 2020